It is generally known that, with the aid of multi-energy CT systems it is possible to record simultaneously CT image data records of an object with different X-ray spectra. As a rule, the attempt is made to calculate from these CT image data records a single image with the desired information, such as a CT image which corresponds to a mono-energy spectrum, or only represents the contrast agent distribution in the scanned object. To this end, a linear-mixed image
      M    m    =            c              0        ,        m              +                  ∑        i            ⁢                        c                      i            ,            m                          ·                  X          i                    can be produced, wherein Xi designates the den CT image data record for the i-th X-ray spectrum, the coefficients are selected such as to achieve a special image impression, and m=0. Unfortunately, the noise in such linear combinations can increase substantially in comparison with a noise-optimized mixed image.
In the prior art there are many possibilities known for reducing the noise of the mixed images which are produced. For example, a linear filter in the form of a low-pass filter can be used, wherein, however, the spatial resolution is sharply reduced. As an alternative, non-linear filters are used, but, with complex structures in the object represented, the problem arises of distinguishing fine structures from the noise.